I opened Excel to demonstrate a formula… and Copilot started writing it for me!
Not a vaguely similar formula. The exact formula I was about to teach.
Quite an interesting moment when you are the person standing at the front of an Excel class to teach formulas.
I recently worked alongside Alpha Development on a multi-part Excel training program for a large investment management company. The program spanned multiple regions, and I built the course material from scratch around the learners’ needs and uses.
Copilot was to be seen as far more than optional sidebar. The client wanted it incorporated part and parcel into the learning experience.
That made this one of the first longer engagements where I had to think seriously about what happens when you teach Excel and Copilot together. It was exciting, occasionally messy, and much more revealing than simply adding a bonus AI module to an existing course.
The central challenge: Copilot makes it easier to create and explore in Excel. It also makes a traditional “follow me and get the same answer” class much harder to run.
First, the good news: some things are stabilizing
I have had earlier conversations with organizations that believed they had paid Copilot licenses, only to discover at the start of a session that the tool was not actually available to the learners.
That was not the case here. Copilot was provisioned and ready to go.
I also did not need to spend a long time on the preliminary questions that once dominated workplace AI conversations: “Does it hallucinate?” “Where does the data go?” “Are we allowed to use it?”
Organizations and learners seem to have a better baseline understanding of them now. That gave us room to focus on a more interesting question:
How should people learn Excel when AI can perform so much of the mechanical work?
Copilot is already an excellent instructor-side tool
From the instructor’s side of the classroom, Copilot is tremendous.
- Need a fresh dataset for another round of practice? It can help produce one.
- Need a data dictionary to make that dataset understandable? It can draft one.
- Need to adapt an example for a different group or business context? That becomes much faster.
- Need another way to explain a formula? Sometimes Copilot explains it better than I do.
I’m only partially joking about that last one. Generative AI is very good at reformulating an explanation, producing variations and helping an instructor adapt in the moment.
But abundance creates its own problem. AI can help produce material at extraordinary speed, but the instructor still has to decide what deserves to make it into the classroom.
You have to be a steward of the material, not a distributor of AI slop.
The learner experience is more complicated
Traditional technical training is designed around consistency. Everyone opens the same file and follows the same sequence. If things go well, everyone arrives at roughly the same result.
Generative AI disrupts that model.
Twenty people can ask Copilot essentially the same question and receive twenty slightly different answers. Suddenly, everyone’s workbook looks different, and the class can turn into chaos.
There is also an engagement paradox. Some learners want to experiment with Copilot and jump straight to the answer. That can make them more skeptical about why they need to learn Excel at all.
At the same time, the wider excitement around AI has made many people more interested in developing their digital skills. It has lit a fire under learners who might previously have treated Excel training as an obligation.
| What Copilot makes easier | What it makes harder |
|---|---|
| Trying an unfamiliar task | Keeping the class on one common path |
| Generating formulas and examples | Knowing whether learners understand the result |
| Offering alternative explanations | Troubleshooting many different answers |
| Reducing intimidation | Preventing learners from skipping the underlying thinking |
| Encouraging experimentation | Fitting Excel skills, prompting and AI judgment into limited class time |
What I had to change
If anything, it would be to be more deliberate about when and why we used Copilot.
1. Separate guided learning from open exploration
There are moments when beginners benefit from following a common path. The class needs shared language and a shared point of reference before everyone branches into a different AI-generated solution.
Copilot works better when exploration is intentional… not when it quietly takes over every exercise.
2. Teach learners to evaluate, not merely generate
If Copilot produces a formula, the learner still needs to ask whether it references the right cells, handles the relevant exceptions and produces a result that makes sense.
The less time learners spend typing a formula, the more important it becomes that they can judge whether that formula deserves to be trusted.
3. Resist teaching every advanced feature at once
Copilot can involve much more than entering a prompt. Personalization, custom instructions, workbook-specific rules and reusable skills can all improve the experience.
But when the learners are Excel beginners and the schedule is limited, introducing all of that can feel like teaching two technical subjects simultaneously. The question is not simply what Copilot can do. It is what this group is ready to absorb today.
4. Build around the business reason
AI can create realistic data and polished examples. But even a beautiful exercise falls flat if learners do not understand why the task matters.
Why would I use this? Where does it show up in real work? What goes wrong if I do it badly?
The best answers often come from lived experience: strange things you have seen in client work, mistakes you have made yourself, and the small decisions that separate a technically correct workbook from one that is genuinely useful.
A practical model for AI-assisted Excel training
- Excel mechanics: Understand what the workbook is doing.
- AI collaboration: Describe the task clearly and iterate productively.
- Judgment: Test the output and connect it to the business problem.
The Excel instructor’s role is changing… not disappearing
This program made me rethink what an Excel trainer is there to do.
Part of the job is still instruction. Learners need a foundation, and beginners cannot reliably evaluate an AI-generated answer without understanding something about the tool underneath it.
But the instructor is also becoming a designer, coach and—in the best sense—a cheerleader. The job is to create an environment where learners are willing to try things, build something of their own, question the first answer and become less intimidated by the blank workbook.
That is a significant change from a technical class where everyone follows the leader and reproduces the same result.
I am looking forward to doing more programs like this because the model is still evolving. We are learning which skills people should memorize, which tasks they can comfortably delegate and what they need to understand in order to use AI responsibly.
Copilot may be able to write the formula. The instructor still has to make people give a damn about why the formula matters.
That is harder to autocomplete.
Is your Excel training ready for Copilot?
Adding Copilot to an existing course is easy. Redesigning the learning experience around it takes more thought.
I help organizations build practical Excel and AI training around the work their teams actually perform—without losing the underlying skills, judgment and confidence that make the tools useful.